Validation of wastewater data using artificial intelligence tools and the evaluation of their performance regarding annotator agreement.

To prevent the pollution of water resources, the measurement and the limitation of wastewater discharges are required. Despite the progress in the field of data acquisition systems, sensors are subject to malfunctions that can bias the evaluation of the pollution flow. It is therefore essential to identify potential anomalies in the data before any use. The objective of this work is to deploy artificial intelligence tools to automate the data validation and to assess the added value of this approach in assisting the validation performed by an operator. To do so, we compare two state-of-the-art anomaly detection algorithms on turbidity data in a sewer network. On the one hand, we conclude that the One-class SVM model is not adapted to the nature of the studied data which is heterogeneous and noisy. The Matrix Profile model, on the other hand, provides promising results with a majority of anomalies detected and a relatively limited number of false positives. By comparing these results to the expert validation, it turns out that the use of the Matrix Profile model objectifies and accelerates the validation task while maintaining the same level of performance compared to the annotator agreement rate between two experts.

Do storm event samples bias the comparison between sewer deposits contribution?

Previous researches demonstrated the occurrence of unique in-sewer sediment, the organic layer, on the Marais site in Paris, capable of explaining the entire wet weather sewer production of suspended particles. Other studies on sites having no similar organic sediment, Clichy in Paris and Ecully in Lyon, demonstrated a wet weather sewer deposits contribution (SDC) to effluent pollution comparable to that of the Marais site, casting therefore doubts on the implication of the organic layer to the outlet discharge pollution. So, an in-depth comparative investigation of the different sites' mean SDC was carried out to confirm or refute the major role of this layer vis-à-vis sewer production. The size and characteristics of the events' sample used to calculate the SDC were analyzed to find whether a statistical bias may have masked a difference that would be more coherent with field observations. After homogenizing these elements, the organic layer regained some of its previously alleged participation in sewer contribution (a maximum of 36% of the total SDC) but another unknown source was still dominant. This suggests that sewer sediment production during wet weather is a result of multiple sediment erosion: the organic layer and another major source not yet identified.

Stochastic evaluation of annual micropollutant loads and their uncertainties in separate storm sewers.

This article describes a stochastic method to calculate the annual pollutant loads and its application over several years at the outlet of three catchments drained by separate storm sewers. A stochastic methodology using Monte Carlo simulations is proposed for assessing annual pollutant load, as well as the associated uncertainties, from a few event sampling campaigns and/or continuous turbidity measurements (representative of the total suspended solids concentration (TSS)). Indeed, in the latter case, the proposed method takes into account the correlation between pollutants and TSS. The developed method was applied to data acquired within the French research project "INOGEV" (innovations for a sustainable management of urban water) at the outlet of three urban catchments drained by separate storm sewers. Ten or so event sampling campaigns for a large range of pollutants (46 pollutants and 2 conventional water quality parameters: TSS and total organic carbon (TOC)) are combined with hundreds of rainfall events for which, at least one among three continuously monitored parameters (rainfall intensity, flow rate, and turbidity) is available. Results obtained for the three catchments show that the annual pollutant loads can be estimated with uncertainties ranging from 10 to 60%, and the added value of turbidity monitoring for lowering the uncertainty is demonstrated. A low inter-annual and inter-site variability of pollutant loads, for many of studied pollutants, is observed with respect to the estimated uncertainties, and can be explained mainly by annual precipitation.